Article on Cell and its growing role at IBM..

[one]

Off-topic for a second, since Cell is created by IBM, Sony, and Toshiba, when military or medical companies buy a Cell-based computer, do Sony, IBM, and Toshiba all get royalties or does just IBM get royalties?

I guess the royalty for the Cell processor is paid to all 3 companies. So, when IBM sell a computer with Cell, they manufacture Cell processors by paying the royalty to the STI consortium, embed them in their computers, then sell them to customers. In other words it'd be no different from other IP-based cores such as ARM, MIPS, SPARC, and Power.

Though Asher seems anxious about the share of IBM in the royalty for the Cell processor, it has the demand of Playstation this year, and then as SPM points out other embedded markets will follow. No other chips manufctured or researched by IBM, including CPUs for servers, supercomputers, Nintendo Revolution and Xbox 360 enjoy this volume of demand except for low-end embedded chips currently manufactured by IBM for products such as printers. But the current low-end market is in intense competition with other suppliers. Cell will tackle the new area of embedded machines, or realtime computers, and IBM can sit doing nothing to receive the huge royalty while Sony/Toshiba are doing the hard work in the embedded market, which sounds good enough for IBM, doesn't it?

 

[Graham]

taken with a bag of salt, it's not really all that bigger news. "Cell and related technologies" is a pretty broard term after all.

I personally don't see cell being the ultimate 'do it all chip' that some may claim. The sun niagra for example (as mentioned) in terms of performance per watt for database applications, I'd be surprised if cell even got close. Simply put a chip cannot excell at every task, especailly one apparently requiring such high tayloring of code to the processor.

Admittadly I'm no expert on cell, but then again I would doubt that anyone else on this board is either (unless under NDA of course). More fuel to the misinformation fire I guess.

 

[one]

taken with a bag of salt, it's not really all that bigger news. "Cell and related technologies" is a pretty broard term after all.

I personally don't see cell being the ultimate 'do it all chip' that some may claim. The sun niagra for example (as mentioned) in terms of performance per watt for database applications, I'd be surprised if cell even got close.

Have you read this section of the news?

Last summer, I spoke with Bijan Divari, who heads the Quasar project. "Five to 10 years from now," he told me, "practically all transactions will be image-based, not text-based. Text is the most painful way of doing transactions."

Without evaluating this forecast your arguments are pointless. Sure, UltraSparc T1, 8 cores @ 1.2Ghz, will be good for backend servers, and will compete with Opteron, POWER and Xeon. But it ends there.

 

[Gubbi]

Have you read this section of the news?

Yeah, I can see processing TIFF screenshots/dumps of XML documents as the natural next step in server software development.

Cheers

 

[Crossbar]

Yeah, I can see processing TIFF screenshots/dumps of XML documents as the natural next step in server software development.

Cheers

It does not take much effort to find an example of image based transaction.

Here's a company working with this already.
http://www.ncr.com/en/solutions/payment_solutions/transaction_manager.htm

And here are some benefits they claim

- Reduced keying labor: ImageMark Transaction Manager uses artificial-intelligence-based software to automatically recognize both handwritten and printed courtesy amounts and account numbers in check and deposit slip images. Reducing manual keying by up to 80 percent!

- Faster transaction balancing: ImageMark Transaction Manager's intelligent, image-based software lets operators balance transactions much more quickly. This software identifies suspected errors, sends the transaction to operators stationed at high-resolution image display terminals, guides the operator through the correction process, and generates advice notices automatically.

But it may be used in a wider term I guess, incorporating biometrics like face recognition etc.

 

[Gubbi]

It does not take much effort to find an example of image based transaction.

Here's a company working with this already.
http://www.ncr.com/en/solutions/payment_solutions/transaction_manager.htm

Which is client side. Nobody in their right mind is going to move this easily distributable OCR software on their server.

And here are some benefits they claim

But it may be used in a wider term I guess, incorporating biometrics like face recognition etc.

I can see image based certification, yes. Image based transactions? No way!

Cheers

 

[Crossbar]

Which is client side. Nobody in their right mind is going to move this easily distributable OCR software on their server.

Why not? It would make sense to me to store scanned images or whatever on a central server why not keep the processing there as well?

I think thin clents are here to stay.

 

[ADEX]

I can see image based certification, yes. Image based transactions? No way!

There's plenty happening already.

What do you thing a bar code is?

 

[Gubbi]

There's plenty happening already.

What do you thing a bar code is?

I think it's a binary number, (modulated and) printed in binary form.

Do you think that the backend of any bar code scanning equipment is handling the raw scanned data?

Cheers

 

[Crossbar]

Image-based transactions on a server do not seem that exotic as they are already used in performance tests:

The performance tests of SYNORAN's eP2: Enterprise Payments Platform in an HP Linux environment demonstrated conclusively that the leading-edge, Java-based system can perform at a sustained transaction rate in excess of 1,300 image-based payments transactions per second and 4,800,000 transactions per hour.

http://www.synoran.com/News_Synoran_Benchmarks_Payments_Technology.htm

This kind of supports that Bijan may see the Cell used primary in IBM servers. I agree with "one" that IBM may leave the embedded market to Sony and Toshiba. After all Kutaragi has already presented "micro cells" with one SPE for smaller devices.

I also agree with Gholbine and MegaDrive1988 that PS4 may likely be built on a scaled version of Cell. That's not obvious if you look at PS1, PS2 and PS3 as they are very different architectures. But I think Sony can't afford changing architecture one more time if they want to maintain backward compatibility because it's just grown too complex. I think that is one reason why Sony has invested "billions" of dollars in the cell, they want an architecture that will scale one more generation without completely changing programing paradigms. That should keep the game developers happy as well.

 

[Carl B]

Well remember that the 'billions' came from fab build-outs, not the R&D specifically - so even if they went with an entirely new architecture, the investment would still be utilized as they will always need fabs for chips. That said though - and like I've said in the past - I'm certain that at this stage in the game, PS4 is looking to be Cell-based. There would have to be a new architecture in the next five years that seriously outperforms Cell on a die-area/performance ratio to imagine Kutaragi and crew switching from Cell, on top of which a lot of the development base they're going to have built around Cell in the community by that time would then be abandoned - a bad position to be in.

I see PS4 as essentially going the route of PS3 only 'more.' Maybe some architectural tweaks to Cell beyond scaling as well, and who knows on the GPU side of things. But if Cell figures in to their long-term network strategy as well - and it seems like it might - so much more will be the penalty to be paid by moving away from it in their next console.

I agree with Asher that there is over-much hype surrounding Cell and it's ultimate potential as reported via certain news/PR outlets, but at the same time I think Cell deserves some respect for the volumes alone it will achieve, and the fact that yes, this architecture will penetrate beyond what the EE ever did. Medical and military may seem quite niche, but they are a nice niche to have recognition in, and if Cell gains long-term traction in these fields along with some supercomputing applications here and there - on top of the bread and butter embedded and PS markets - I will think that the entire Cell inititiative would qualify as a success for sure. It's not 'Cell taking over the world' or any madness like that, but it'd be a quite respectable performance by the architecture all the same.

 

[koldfuzion]

Interesting thread guys--and thanks Asher for the reality check--I'm sure IBM has plenty of internally competing silos. However, I agree with XB that cell will be implemented in a range of apps, including defense, servers, medical devices, PS3, nearly every Toshiba TV and others. I agree that it will not be in "every" IBM machine made (that is clearly puffery) but I do they intend to maximize operating leverage over the very large fixed capital base and R&D investment--some apps I do not think will be niche at all--some will be in mass-market products, TVs are already in the bag. Anyway this is not necessarily a huge deal, how many people care what chip is in their cell phone--eg I had no idea Nvidia powered my cell phone until I read Nvidia's 10-Q. But I do think it will be interesting to see how the business model develops over time for this architecture. Hyped, yeah sure, but I don't think it's going to be some backwater architecture either.

 

[3dcgi]

Have you read this section of the news?
Without evaluating this forecast your arguments are pointless. Sure, UltraSparc T1, 8 cores @ 1.2Ghz, will be good for backend servers, and will compete with Opteron, POWER and Xeon. But it ends there.

Add some floating point units to Niagra (Rock) and beef up the clock speed and you've got the theoretical power of Cell, more or less. Then it comes down to the threading/memory model. It's too early to declare a winner here.

 

[one]

Add some floating point units to Niagra (Rock) and beef up the clock speed and you've got the theoretical power of Cell, more or less. Then it comes down to the threading/memory model. It's too early to declare a winner here.

Of course technically it will be able to reach certain performance with its outstanding throughput though I doubt it can get a very high clock-speed (it consumes 72W at 1.2Ghz). But, as Asher points out about IBM as an organizational entity with inner turf battles, we have to think in terms of the realistic economy. Which chip project can get the biggest investment? A project with the prospect of the biggest profit.

Same for Sun's Niagra and its beefed-up brother Rock in 2008, will they have the demand of Playstation that justifies floating-point oriented redesign and bigger investment?

 

[3dcgi]

I agree scaling clock speed without blowing the power budget will be hard and I don't see anything outside of x86 rivaling the investment in Cell. Sun definitely has an uphill battle ahead of them and STI seems to be thinking in more grandiose terms so we'll see if that helps them in the long run.